CN111985611A - Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution - Google Patents
Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution Download PDFInfo
- Publication number
- CN111985611A CN111985611A CN202010705465.2A CN202010705465A CN111985611A CN 111985611 A CN111985611 A CN 111985611A CN 202010705465 A CN202010705465 A CN 202010705465A CN 111985611 A CN111985611 A CN 111985611A
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- network model
- training
- convolutional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000001259 photo etching Methods 0.000 title claims abstract description 49
- 238000010801 machine learning Methods 0.000 title claims abstract description 39
- 238000010586 diagram Methods 0.000 title claims abstract description 26
- 238000004364 calculation method Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 94
- 238000003062 neural network model Methods 0.000 claims abstract description 83
- 238000001459 lithography Methods 0.000 claims abstract description 72
- 238000003384 imaging method Methods 0.000 claims abstract description 71
- 238000010606 normalization Methods 0.000 claims abstract description 66
- 230000003287 optical effect Effects 0.000 claims abstract description 44
- 238000012795 verification Methods 0.000 claims abstract description 38
- 238000005070 sampling Methods 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims description 64
- 238000000034 method Methods 0.000 claims description 46
- 230000001427 coherent effect Effects 0.000 claims description 30
- 238000013461 design Methods 0.000 claims description 28
- 238000005286 illumination Methods 0.000 claims description 18
- 238000000354 decomposition reaction Methods 0.000 claims description 10
- 238000013527 convolutional neural network Methods 0.000 description 16
- 230000008569 process Effects 0.000 description 13
- 238000005516 engineering process Methods 0.000 description 11
- 238000013507 mapping Methods 0.000 description 10
- 239000004065 semiconductor Substances 0.000 description 6
- 239000013598 vector Substances 0.000 description 5
- 238000004891 communication Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000001900 extreme ultraviolet lithography Methods 0.000 description 2
- 238000000671 immersion lithography Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007654 immersion Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012634 optical imaging Methods 0.000 description 1
- 238000000059 patterning Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000000206 photolithography Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 210000001747 pupil Anatomy 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005316 response function Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L21/00—Processes or apparatus adapted for the manufacture or treatment of semiconductor or solid state devices or of parts thereof
- H01L21/02—Manufacture or treatment of semiconductor devices or of parts thereof
- H01L21/027—Making masks on semiconductor bodies for further photolithographic processing not provided for in group H01L21/18 or H01L21/34
- H01L21/0271—Making masks on semiconductor bodies for further photolithographic processing not provided for in group H01L21/18 or H01L21/34 comprising organic layers
- H01L21/0273—Making masks on semiconductor bodies for further photolithographic processing not provided for in group H01L21/18 or H01L21/34 comprising organic layers characterised by the treatment of photoresist layers
- H01L21/0274—Photolithographic processes
Abstract
A calculation method based on physical imaging characteristic diagram and DCNN machine learning reverse lithography comprises the step of basing the lithography target pattern on an optical scale Ki(x, y) obtaining a feature atlas of N1 { Si }; establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the input of the neural network model is a feature atlas of { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training; during training of the neural network model, different combinations of input end channel dimensions, different convolutional layer and batch standardization layer numbers N2 and different convolutional kernel numbers N3 of each convolutional layer and batch standardization layer are adopted for training and verification until a satisfactory combination of the neural network model is foundMolding; inputting the feature atlas of N1 { Si } of the wafer photoetching target pattern into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
Description
Technical Field
The invention belongs to the field of integrated circuit manufacturing, and provides a machine learning reverse photoetching solution calculation method based on a physical imaging characteristic diagram and a Deep Convolutional Neural Network (DCNN).
Background
Computational lithography plays a vital role in the semiconductor industry. When the semiconductor technology node is reduced to 14nm or below, the photoetching technology is gradually close to the physical limit, and light Source Mask Optimization (SMO for short) is used as a novel resolution enhancement technology, so that the overlapping process window of semiconductor photoetching under the limit size can be obviously improved, and the life cycle of the conventional photoetching technology can be effectively extended. SMO is not only an important component of 193nm immersion lithography, but will also be an essential technology in EUV lithography.
As the semiconductor industry moves toward technology nodes above 7nm, the gap between the performance control capability of lithography machines and the control requirements for lithography process edge placement errors is rapidly expanding. To enable the semiconductor industry to continue to advance with the use of current lithography machines (i.e., immersion scanners or EUV scanners), computational lithography continues to play a critical role. The method is used as a novel resolution enhancement technology in Source Mask Optimization (SMO for short), can remarkably improve the overlapping process window of semiconductor lithography under the limit size, and effectively extends the life cycle of the conventional lithography technology at present.
Then, Inverse Lithography Technology (ILT) becomes the final frontier in the field of computational Lithography. Inverse lithography aims at calculating the optimal mask pattern to achieve the target pattern required for imaging on the wafer, and is maximally robust to process variations. At present, various strict reverse photoetching algorithms exist, but because of huge requirements of computing hardware resources and too long computing time, the realization of the strict reverse photoetching on a whole chip is still unrealistic. Moreover, the 3D effect of EUV masks is more pronounced, and full-chip strictly inverse lithography for EUV is particularly difficult to achieve.
Referring to FIG. 1, FIG. 1 is an idealized schematic diagram of a process from patterning to rigorous reverse lithography. As shown in FIG. 1, the left side of the figure is the lithographic target design pattern and the right side is the exact inverse lithographic solution of the mask. It is clear to those skilled in the art that all machine learning based computational lithography techniques, including machine learning based inverse lithography techniques, need to solve the problem of how to characterize the environment around a point, where the response of a certain point (x, y) depends only on the neighboring environment within its influence range, which is essentially a feature vector design.
This problem can be expressed as:
as can be seen from the above, the ILT solves the mapping that is not point-to-point, but rather from a function to a point. The value of the ILT solution at point (x, y) depends not only on the value of the lithographic target design pattern at point (x, y), but also on all values of all lithographic target design patterns within the optical influence range. The optical influence range is usually around 1.5 μm to 2.0. mu.m.
Machine learning techniques based on neural network architecture are a very promising technique to overcome the practical application obstacles of Inverse Lithography (ILT). Machine learning based ILT has two key components:
extracting or designing (information coding) a feature vector;
and (II) designing a neural network structure constructed by the mapping function.
That is, the eigenvector design is essentially a scheme that encodes or measures the neighboring environment around any point (x, y) of the lithographic target design pattern.
Important criteria for feature vector design are measurement resolution, measurement sufficiency (completeness), and measurement validity. The design of the feature vectors also has an important influence on the structure of the mapping function. To better implement machine learning-based ILT model generalization, the mapping function should achieve the following ideal properties:
the curvature should be as small as possible;
II, monotonous;
thirdly, smooth or differentiable.
The solution to Inverse Lithography (ILT) is obtained using a Deep Convolutional Neural Network (DCNN), and it has been reported in the literature that in prior art implementations, the input to the neural network model is the lithographic target design pattern. Thus, the input to the model is binary in nature, and the mapping function is a continuous function from a binary function with unlimited bandwidth to a bandwidth limited by the strict ILT algorithm. Thus, there are two essential disadvantages to using lithographic target design patterns directly as inputs to Deep Convolutional Neural Networks (DCNN):
measuring the environment around a point (x, y) is on a purely geometric scale. However, the lithographically formed image is not simply formed of pure geometry, and the optical properties of the geometry and background (light transmission and phase variation) must be considered;
more importantly, the imaging process under the illumination condition of the partially coherent light is a nonlinear process. Therefore, the adoption of the pure geometric pattern of the photoetching target design pattern as the DCNN input has intrinsic defects in nature, which causes a feature vector extraction layer in the DCNN model to be more complicated;
the mapping function from "pure geometric space" to the ILT solution space is not a smooth function, and there are many places where the curvature is discontinuous (as shown in fig. 2).
Disclosure of Invention
In order to overcome the defects of the existing DCNN-ILT implementation, the invention provides a machine learning reverse photoetching calculation method based on a physical imaging characteristic diagram and a deep convolutional neural network, which uses the information of a wafer imaging end as input and is combined with the specially designed deep convolutional neural network to realize high-precision machine learning-based reverse photoetching solution.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a calculation method based on a physical imaging characteristic diagram and DCNN machine learning reverse lithography solution is used for predicting/calculating a value of the reverse lithography solution; the method comprises the following steps:
step S11: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S12: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining a feature atlas of N1 { Si };
step S13: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the channel dimension of the input layer is equal to N1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolutional cores of each convolutional layer and batch normalization layer is N3, and N2 and N3 are positive integers;
step S14: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the feature atlas of the { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S15: during neural network model training, training with the training sample by using the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolution and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
Further, the method also includes step S16: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si }, and inputting the feature map set of N1 { Si } into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
In order to achieve the above object, another technical solution of the present invention is as follows:
a calculation method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution is used for predicting/calculating the value of the reverse photoetching solution; characterized in that the method comprises the following steps:
step S21: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S22: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S23: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the channel dimension N1 of the input layer is 1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolutional cores of each convolutional layer and batch normalization layer is N3, and N2 and N3 are positive integers;
step S24: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the imaging light intensity function; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S25: during neural network model training, training with the training sample by using the input end channel dimension, N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification. Further, the method further includes step S26: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si } and an imaging light intensity function of the feature maps of the N1 { Si } under partially coherent illumination, and inputting the imaging light intensity function into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a calculation method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution is used for predicting/calculating the value of the reverse photoetching solution; the method comprises the following steps:
step S31: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S32: the light is emittedThe engraved target pattern is based on the optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S33: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the dimension of the input layer is N1+1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolutional cores of each convolutional layer and batch normalization layer is N3, and N2 and N3 are positive integers;
step S34: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the combination of an imaging light intensity function and a feature map set of { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S35: during neural network model training, using different combinations of the input end channel dimension, different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3, training with the training sample, and verifying with a verification sample until finding the neural network model with a satisfactory combination of the input end channel dimension, the number of different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
Further, the method further includes step S36: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si }, the feature map of N1 { Si } having an imaging intensity function under partially coherent illumination, anAnd inputting the combination of the imaging light intensity function and the feature atlas of the { Si } into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
Compared with the prior ILT technology based on machine learning, the invention does not use the photoetching target pattern as the input of the DCNN model, but uses the photoetching light intensity image in the photoetching process, or the characteristic signal image set, or the combination of the photoetching light intensity image and the characteristic signal image as the input of the DCNN model, and combines with the specially designed deep convolution neural network, thereby realizing the high-precision reverse photoetching solution based on machine learning; also, the method is not limited to immersion lithography, and it is equally applicable to EUV lithography.
Drawings
FIG. 1 is a schematic diagram of a deep neural network in the prior art
FIG. 2 is a diagram illustrating a mapping function from "pure geometric space" to ILT solution space in the prior art
FIG. 3 is a schematic diagram of the whole imaging process under the illumination of the partially coherent light in the embodiment of the present invention
FIG. 4 is a schematic diagram of the architecture of DCNN machine learning reverse lithography based on physical imaging feature mapping in an embodiment of the present invention
FIG. 5 is a flowchart illustrating a computing method based on a physical imaging profile and DCNN machine learning inverse lithography solution according to an embodiment of the present invention
FIG. 6 is a schematic diagram of the reverse lithography results (feature signal image set { S1, S2, S3, S4, S5} as input to DCNN) of machine learning based on physical imaging feature maps in an embodiment of the present invention
FIG. 7 is a schematic diagram of the reverse lithography result (lithography light intensity image as input to DCNN) based on machine learning of physical imaging profiles in an embodiment of the present invention
FIG. 8 is a schematic diagram of the reverse lithography result (lithography light intensity image + feature signal image set { S1, S2, S3, S4, S5} as input to DCNN) of machine learning based on physical imaging feature map in an embodiment of the present invention
Detailed Description
The following description of the present invention will be made in detail with reference to the accompanying drawings 1 to 8.
The invention discloses a calculation method based on a physical imaging characteristic diagram and DCNN machine learning reverse lithography. In order to overcome the defects of the existing DCNN-ILT realization, the method provides that the information of a wafer imaging end is used as input and is combined with a specially designed deep convolution neural network, and the high-precision reverse photoetching solution based on machine learning is realized.
The design principle of the invention is inspired by imaging physics, and it is clear to those skilled in the art that, based on the imaging theory, the calculation of the imaging light intensity function under the partially coherent illumination by using the optimal coherent decomposition method can be calculated and embodied by the following equations (2a) and (2 b):
wherein the content of the first and second substances,
representing a function KiConvolution operation between (x, y) and mask transfer function M (x, y). { alpha ]iAnd { K }iThe eigenvalues and eigenfunctions of the following equations.
∫∫W(x1',y1'；x2',y2')Ki(x2',y2')dx2'dy2'＝αiKi(x1',y1') (3a)
W(x1',y1'；x2',y2')＝γ(x2'-x1',y2'-y1')P(x1',y1')P*(x2',y2') (3b)
In the formula, gamma (x)2-x1Y2-y1) is (x) in the object plane (i.e. the mask plane)1,y1) And (x)2,y2) The mutual coherence factor between them, determined by the illumination conditions; p (x-x)1,y-y1) Is the impulse response function of the optical imaging system and is determined by the pupil function of the optical system. More specifically, it is due to a point (x) in the object plane1,y1) The unit amplitude and zero phase source at (x, y) is the complex amplitude at point (x, y) in the image plane.
The significance of equations (2a) and (2b) above is that it indicates that partially coherent imaging systems can be decomposed into a series of coherent imaging systems, and that the written coherent imaging systems are independent of each other. The overall imaging process under partially coherent illumination can be described in information by the following figure 3.
Referring to fig. 3, fig. 3 is a schematic diagram illustrating the whole imaging process under the illumination of the partially coherent light according to the embodiment of the present invention. As shown in fig. 3, there are the following important observation features from the viewpoint of the imaging process:
in Fourier space, reverse photoetching only needs to optimize bandwidth NA (1+ sigma)max) A frequency order within;
the lithographic image is obtained by projecting the frequency order onto the communication kernel of a series of imaging systems;
only information that can pass through these communication channels is useful for reverse photolithography.
From the above, it can be seen that the imaging system communications core is a natural set of "optical scales" that measure or estimate the environment around a point (x, y). In the embodiment of the present invention, the { Si } feature map calculated by equation (2b) or the imaging light intensity function calculated by equation (2a) may be used; or a combination of the imaging intensity function and the { Si } profile as inputs to the DCNN structure.
Specifically, referring to fig. 4, fig. 4 is a schematic diagram illustrating an architecture of DCNN machine learning reverse lithography based on physical imaging feature mapping according to an embodiment of the present invention. As shown in fig. 4, the characteristic map of the DCNN model input may be a calculated { lithography intensity function image }, or a { signal image set, S1(x, y), S2(x, y), … SN1(x, y) }, or a { lithography intensity function image + signal image set, S1(x, y), S2(x, y), … SN1(x, y) }.
The advantage of using the three signatures described above as inputs to the DCNN model is that they take full advantage of the imaging process and physical characteristics, i.e., the physics of light transmission and phase changes that take into account the lithographic design geometry and its background, as well as the physical characteristics of the image under a given illumination condition. Like the strict ILT solution function, the feature map image is a continuous frequency bandwidth limited function, and therefore, the design can significantly reduce the non-linearity of the mapping function.
In the embodiment of the present invention, in order to further improve the efficiency of adjacent environment coding, a special deep convolutional neural network is developed, and the deep convolutional neural network structure needs to have the following good performance:
the sensing visual field is as wide as possible, and the wider the sensing visual field is, the better the capability of exploring spatial information around a certain point (x, y) is;
the original resolution of the image should remain unchanged;
the depth of the deep convolutional neural network should be moderate, so that residual connection does not need to be added in the network structure, and the deep convolutional neural network training is facilitated.
In accordance with the above design guidelines, in embodiments of the present invention, all pooling layers may be replaced with normalization layers, and ReLu is used as the activation function, with convolution kernels each having a size of 3x3 and a step size of 1.
The training of the deep convolutional neural network needs to include training samples and test samples. For example, samples may be taken from the peripheral control circuit area of a 28nm SRAM design. The total number of images used for training was 134, and the total number of images used for the above deep convolutional neural network model test was 48. In the deep convolutional neural network training, He weight initialization and orthogonal initialization are tried, and the two different weight initialization schemes are found to have no essential difference in the quality of the deep convolutional neural network. Make itThe learning rate used may be 5x10-5Furthermore, the deep convolutional neural network model described above can be optimized using an Adam optimizer in the training.
Furthermore, to assess the quality of the deep convolutional neural network described above, first, a strict inverse lithography solution can be normalized to [0,1 ] using a common normalization factor]Then, two metrics are used to quantify the quality of the model. Expressing the normalized strictly inverse photolithographically resolved image by O
Representing a neural network model predicted image. The first metric used is probability
Where 0.1 and 0.05, another metric is RMSE, which equals
| … | represents the Frobenius norm, n is the total number of pixels, and the model training error statistics and the trial error statistics are shown in table 1 below.
The analysis image is a space imaging light intensity distribution image, the Model input is a Model input, the Error spec is an Error standard, the Training set is a Training sample set, and the Test set is a verification sample set.
The following describes in detail a calculation method based on a physical characteristic diagram and DCNN machine learning inverse lithography solution according to the present invention by three embodiments.
Example 1
Referring to fig. 5, fig. 5 is a flowchart illustrating a calculation method based on a physical imaging profile and DCNN machine learning reverse lithography according to an embodiment of the present invention. In the embodiment of the present invention, the method for calculating a reverse lithography solution based on a physical feature map and DCNN machine learning is used for predicting/calculating a value of the reverse lithography solution, and includes the following steps:
step S11: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer; wherein N1 is a positive integer;
step S12: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining a feature atlas of N1 { Si };
step S13: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; wherein the channel dimension of the input layer is equal to N1, the convolutional layers and batch normalization layers have N2 layers (depth of convolutional network), and the number of convolutional cores of each convolutional layer and batch normalization layer is N3 (convolutional layer width); wherein N2 and N3 are positive integers.
In the design of the neural network, the width of all convolutional layers is the same. Assuming that each feature image dimension is 2p × 2q × 1, the down-sampling layer converts the feature image dimension to p × q × 4, i.e., decreases the dimension on the xy plane while increasing the number of channels. The upsampling layer dimension is p × q × 4 and the output layer dimension is 2p × 2q × 1.
Step S14: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the feature atlas of the { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S15: during neural network model training, training with the training sample by using the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolution and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
Based on the neural network model, in the application implementation stage, step S16 may be performed on the designed wafer lithography target pattern, i.e. the designed wafer lithography target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si }, and inputting the feature map set of N1 { Si } into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
In the embodiment of the invention, the optical scale K is usediFor example, if the number of (x, y) is 5, the value of the predicted reverse lithography solution can be obtained by inputting the feature signal image set { S1, S2, S3, S4, S5} into the first DCNN model. Referring to fig. 6, fig. 6 is a schematic diagram of the reverse lithography result of machine learning based on physical imaging feature maps (feature signal image set { S1, S2, S3, S4, S5} as input to DCNN) according to an embodiment of the present invention. As shown in fig. 6, comparing the image of the rigorous ILT solution with the machine learning model ILT solution shows that the results are very desirable.
Example 2
In another preferred embodiment of the present invention, a method for calculating an inverse lithography solution based on a physical feature map and DCNN machine learning is used for predicting/calculating a value of the inverse lithography solution, and comprises the following steps:
step S21: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S22: basing the lithographic target pattern onThe optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S23: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the channel dimension of the input layer is 1, the convolutional layers and the batch normalization layer have N2 layers in total, and the number of convolution kernels of each convolutional layer and batch normalization layer is N3 (convolutional layer width); wherein N2 and N3 are positive integers.
In the design of the neural network, the width of all convolutional layers is the same. Assuming that each feature image dimension is 2p × 2q × 1, the down-sampling layer converts the feature image dimension to p × q × 4, i.e., decreases the dimension on the xy plane while increasing the number of channels. The upsampling layer dimension is p × q × 4 and the output layer dimension is 2p × 2q × 1.
Step S24: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the imaging light intensity function; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S25: during neural network model training, training with the training sample by using the input end channel dimension, N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification, where a channel dimension N1 at this time is 1.
Based on the second DCNN model, in the application implementation phase, step S16 may be performed on the designed wafer lithography target pattern, i.e. the designed wafer lithography target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si }, wherein the imaging light intensity function of the feature maps of N1 { Si } under the partially coherent illumination is input into a trained neural network model, and a pattern for predicting reverse photoetching is obtained.
In an embodiment of the invention, the value of the predicted inverse lithography solution may be obtained by applying an imaging light intensity function to the second DCNN model. Referring to fig. 7, fig. 7 is a schematic diagram illustrating the reverse lithography result (lithography light intensity image as input of DCNN) of machine learning based on physical feature map according to an embodiment of the present invention. As shown in fig. 7, the result is more desirable by comparing the image of the rigorous ILT solution with the machine learning model ILT solution.
Example 3
In another preferred embodiment of the present invention, a method for predicting/calculating an inverse lithography solution based on a physical imaging profile and DCNN machine learning inverse lithography solution comprises the following steps:
step S31: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S32: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S33: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the dimension of the input layer is N1+1, the convolutional layers and the batch normalization layer have N2 layers in total, and the number of convolutional cores of each convolutional layer and batch normalization layer is N3 (convolutional layer width); wherein N2 and N3 are positive integers.
In the design of the neural network, the width of all convolutional layers is the same. Assuming that each feature image dimension is 2p × 2q × 1, the down-sampling layer converts the feature image dimension to p × q × 4, i.e., decreases the dimension on the xy plane while increasing the number of channels. The upsampling layer dimension is p × q × 4 and the output layer dimension is 2p × 2q × 1.
Step S34: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the combination of an imaging light intensity function and a feature map set of { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S35: during neural network model training, using different combinations of the input end channel dimension, different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3, training with the training sample, and verifying with a verification sample until finding the neural network model with a satisfactory combination of the input end channel dimension, the number of different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
Based on the neural network model, in the application implementation stage, step S36 may be performed on the designed wafer lithography target pattern, i.e. the designed wafer lithography target pattern is based on the optical scale Ki(x, y) to obtain a set of N1 { Si } signatures, the N1 { Si } signatures being light that is imaged under partially coherent illuminationAnd (3) a strong function, and inputting the combination of the imaging light intensity function and the feature atlas of the { Si } into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
In the embodiment of the invention, the optical scale K is usediFor example, if the number of (x, y) is 5, the predicted reverse lithography solution value can be obtained by inputting the lithography light intensity image + feature signal image set { S1, S2, S3, S4, S5} into the third DCNN model. Referring to fig. 8, fig. 8 is a schematic diagram of the reverse lithography result (lithography light intensity image + feature signal image set { S1, S2, S3, S4, S5} as input to DCNN) based on machine learning of the physical imaging feature map according to an embodiment of the present invention. As shown in fig. 8, comparing the image of the rigorous ILT solution with the machine learning model ILT solution shows that the results are very desirable.
In summary, the present invention fully utilizes the mature physical imaging-based feature map and DCNN machine learning techniques, and utilizes the Deep Convolutional Neural Network (DCNN), so as to obtain the solution of the Inverse Lithography Technology (ILT), which is much faster (about 25 times faster) than the strict inverse lithography calculation.
The above description is only for the preferred embodiment of the present invention, and the embodiment is not intended to limit the scope of the present invention, so that all the equivalent structural changes made by using the contents of the description and the drawings of the present invention should be included in the scope of the present invention.
Claims (6)
1. A calculation method based on a physical imaging characteristic diagram and DCNN machine learning reverse lithography solution is used for predicting/calculating a value of the reverse lithography solution; characterized in that the method comprises the following steps:
step S11: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S12: basing the lithographic target pattern on the optical scale Ki(x, y) isTo a set of N1 { Si };
step S13: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the channel dimension of the input layer is equal to N1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolutional cores of each convolutional layer and batch normalization layer is N3, and N2 and N3 are positive integers;
step S14: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the feature atlas of the { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S15: during neural network model training, training with the training sample by using the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolution kernels of each convolution and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
2. The method according to claim 1, further comprising step S16: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si }, and inputting the feature map set of N1 { Si } into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
3. A calculation method based on a physical imaging characteristic diagram and DCNN machine learning reverse lithography solution is used for predicting/calculating a value of the reverse lithography solution; characterized in that the method comprises the following steps:
step S21: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S22: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S23: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the channel dimension of the input layer is 1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolution kernels of each convolutional layer and each batch normalization layer is N3, and N2 and N3 are positive integers;
step S24: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the imaging light intensity function; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S25: during neural network model training, training with the training sample by using the input end channel dimension, N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer, and verifying by using a verification sample until the neural network model with a satisfactory combination of the input end channel dimension, the number N2 of different convolutional layers and batch normalization layers and the number N3 of convolutional kernels of each convolutional layer and batch normalization layer is found; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
4. The method according to claim 3, further comprising step S26: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining a feature map set of N1 { Si } and an imaging light intensity function of the feature maps of the N1 { Si } under partially coherent illumination, and inputting the imaging light intensity function into a trained neural network model to obtain a pattern for predicting reverse photoetching solution.
5. A calculation method based on a physical imaging characteristic diagram and DCNN machine learning reverse lithography solution is used for predicting/calculating a value of the reverse lithography solution; characterized in that the method comprises the following steps:
step S31: according to imaging conditions, based on an optimal coherent decomposition method, N1 feature function sets { K ] are calculatedi(x, y) }, i ═ 1,2, … N1; wherein the set of feature functions { K }i(x, y) } is an optimal set of optical scales, and N1 is the optical scale KiThe number of (x, y), N1 is a positive integer;
step S32: basing the lithographic target pattern on the optical scale Ki(x, y) obtaining N1 { Si } feature maps and an imaging intensity function of the N1 { Si } feature maps under partially coherent illumination;
step S33: establishing a neural network model, wherein the neural network model comprises an input layer, a down-sampling layer, a convolution layer, a batch normalization layer, an up-sampling layer and an output layer; the dimension of the input layer is N1+1, the convolutional layers and the batch normalization layer have N2 layers in total, the number of convolutional cores of each convolutional layer and batch normalization layer is N3, and N2 and N3 are positive integers;
step S34: training the neural network model by adopting a training sample set and a verification sample set, wherein the training sample set and the verification sample set are part of target patterns in the design target patterns selected randomly; the input of the input end of the neural network model is the combination of an imaging light intensity function and a feature map set of { Si }; generating an optimal mask image by using a strict reverse photoetching algorithm, and taking the optimal mask image as a training target image for neural network training;
step S35: during neural network model training, using different combinations of the input end channel dimension, different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3, training with the training sample, and verifying with a verification sample until finding the neural network model with a satisfactory combination of the input end channel dimension, the number of different convolutional layer and batch normalization layer N2 and the number of convolutional kernels of each convolutional layer and batch normalization layer N3; wherein the satisfactory combination means that, for a predetermined proportion of pixel units of each image in the training set and the verification set, an error between a predicted value of the neural network model and a value of the rigorous inverse lithography solution is less than or equal to a predefined error specification.
6. The method according to claim 5, further comprising step S36: in the application implementation stage, the designed wafer photoetching target pattern is based on the optical scale Ki(x, y) obtaining N1 feature maps of { Si }, inputting the imaging light intensity function of the N1 feature maps of { Si } under partially coherent illumination into a trained neural network model, and inputting the combination of the imaging light intensity function and the feature maps of { Si } into the trained neural network model to obtain a pattern for predicting reverse photoetching solution.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010705465.2A CN111985611A (en) | 2020-07-21 | 2020-07-21 | Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution |
| PCT/CN2020/138332 WO2022016802A1 (en) | 2020-07-21 | 2020-12-22 | Physical feature map- and dcnn-based computation method for machine learning-based inverse lithography technology solution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010705465.2A CN111985611A (en) | 2020-07-21 | 2020-07-21 | Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111985611A true CN111985611A (en) | 2020-11-24 |
Family
ID=73439536
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010705465.2A Withdrawn CN111985611A (en) | 2020-07-21 | 2020-07-21 | Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111985611A (en) |
| WO (1) | WO2022016802A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561873A (en) * | 2020-12-11 | 2021-03-26 | 上海集成电路装备材料产业创新中心有限公司 | CDSEM image virtual measurement method based on machine learning |
| CN112578646A (en) * | 2020-12-11 | 2021-03-30 | 上海集成电路装备材料产业创新中心有限公司 | Offline photoetching process stability control method based on image |
| WO2022016802A1 (en) * | 2020-07-21 | 2022-01-27 | 上海集成电路研发中心有限公司 | Physical feature map- and dcnn-based computation method for machine learning-based inverse lithography technology solution |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115294381B (en) * | 2022-05-06 | 2023-06-30 | 兰州理工大学 | Small sample image classification method and device based on feature migration and orthogonal prior |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180095359A1 (en) * | 2016-09-30 | 2018-04-05 | Samsung Electronics Co., Ltd. | Method of optimizing a mask using pixel-based learning and method for manufacturing a semiconductor device using an optimized mask |
| CN107908071A (en) * | 2017-11-28 | 2018-04-13 | 上海集成电路研发中心有限公司 | A kind of optical adjacent correction method based on neural network model |
| CN108665060A (en) * | 2018-06-12 | 2018-10-16 | 上海集成电路研发中心有限公司 | A kind of integrated neural network for calculating photoetching |
| CN109976087A (en) * | 2017-12-27 | 2019-07-05 | 上海集成电路研发中心有限公司 | The generation method of mask pattern model and the optimization method of mask pattern |
| CN111310407A (en) * | 2020-02-10 | 2020-06-19 | 上海集成电路研发中心有限公司 | Method for designing optimal feature vector of reverse photoetching based on machine learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111985611A (en) * | 2020-07-21 | 2020-11-24 | 上海集成电路研发中心有限公司 | Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution |
-
2020
- 2020-07-21 CN CN202010705465.2A patent/CN111985611A/en not_active Withdrawn
- 2020-12-22 WO PCT/CN2020/138332 patent/WO2022016802A1/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180095359A1 (en) * | 2016-09-30 | 2018-04-05 | Samsung Electronics Co., Ltd. | Method of optimizing a mask using pixel-based learning and method for manufacturing a semiconductor device using an optimized mask |
| CN107908071A (en) * | 2017-11-28 | 2018-04-13 | 上海集成电路研发中心有限公司 | A kind of optical adjacent correction method based on neural network model |
| CN109976087A (en) * | 2017-12-27 | 2019-07-05 | 上海集成电路研发中心有限公司 | The generation method of mask pattern model and the optimization method of mask pattern |
| CN108665060A (en) * | 2018-06-12 | 2018-10-16 | 上海集成电路研发中心有限公司 | A kind of integrated neural network for calculating photoetching |
| CN111310407A (en) * | 2020-02-10 | 2020-06-19 | 上海集成电路研发中心有限公司 | Method for designing optimal feature vector of reverse photoetching based on machine learning |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022016802A1 (en) * | 2020-07-21 | 2022-01-27 | 上海集成电路研发中心有限公司 | Physical feature map- and dcnn-based computation method for machine learning-based inverse lithography technology solution |
| CN112561873A (en) * | 2020-12-11 | 2021-03-26 | 上海集成电路装备材料产业创新中心有限公司 | CDSEM image virtual measurement method based on machine learning |
| CN112578646A (en) * | 2020-12-11 | 2021-03-30 | 上海集成电路装备材料产业创新中心有限公司 | Offline photoetching process stability control method based on image |
| WO2022121736A1 (en) * | 2020-12-11 | 2022-06-16 | 上海集成电路装备材料产业创新中心有限公司 | Cdsem image virtual measurement method based on machine learning |
| CN112578646B (en) * | 2020-12-11 | 2022-10-14 | 上海集成电路装备材料产业创新中心有限公司 | Offline photoetching process stability control method based on image |
| CN112561873B (en) * | 2020-12-11 | 2022-11-25 | 上海集成电路装备材料产业创新中心有限公司 | CDSEM image virtual measurement method based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022016802A1 (en) | 2022-01-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111985611A (en) | Computing method based on physical characteristic diagram and DCNN machine learning reverse photoetching solution | |
| CN107797391B (en) | Optical proximity correction method | |
| US11119401B2 (en) | Simulating near field image in optical lithography | |
| CN107908071B (en) | Optical proximity correction method based on neural network model | |
| CN111627799B (en) | Method for manufacturing semiconductor element | |
| CN109976087B (en) | Method for generating mask pattern model and method for optimizing mask pattern | |
| CN110426914B (en) | Correction method of sub-resolution auxiliary graph and electronic equipment | |
| CN108228981B (en) | OPC model generation method based on neural network and experimental pattern prediction method | |
| CN113238460B (en) | Deep learning-based optical proximity correction method for extreme ultraviolet | |
| CN111310407A (en) | Method for designing optimal feature vector of reverse photoetching based on machine learning | |
| CN108490735A (en) | The method, apparatus and computer-readable medium that full chip mask pattern generates | |
| US7328424B2 (en) | Method for determining a matrix of transmission cross coefficients in an optical proximity correction of mask layouts | |
| US11762283B2 (en) | Inverse lithography and machine learning for mask synthesis | |
| WO2023241267A1 (en) | Training method and apparatus for lithographic-mask generation model, and device and storage medium | |
| CN113495435B (en) | Digital mask projection lithography optimization method and system | |
| WO2022082692A1 (en) | Lithography hotspot detection method and apparatus, and storage medium and device | |
| CN103365071A (en) | Optical proximity correction method for mask plate | |
| US10571799B1 (en) | Hessian-free calculation of product of Hessian matrix and vector for lithography optimization | |
| CN107479335A (en) | The optical imagery quick calculation method decomposed based on light source interaural crosscorrelation function | |
| CN115222829A (en) | Mask optimization method and device based on neural network model | |
| CN117313642A (en) | Training method, device, equipment and storage medium for lithography mask generation model | |
| CN114418070A (en) | Convolution processing method and device | |
| Lin et al. | Compressive sensing method for EUV source optimization using different bases | |
| CN107942614B (en) | Method for generating auxiliary pattern of hole layer and method for generating correction function | |
| CN114898169B (en) | Deep learning-based photoetching OPC database establishment method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20201203 Address after: 201210, Zhangjiang Road, Zhangjiang, Pudong New Area, No. 497, Gauss Road, Shanghai Applicant after: SHANGHAI IC R & D CENTER Co.,Ltd. Applicant after: Shanghai IC equipment Material Industry Innovation Center Co.,Ltd. Address before: 201210, Zhangjiang Road, Zhangjiang, Pudong New Area, No. 497, Gauss Road, Shanghai Applicant before: SHANGHAI IC R & D CENTER Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20201124 |
|
| WW01 | Invention patent application withdrawn after publication |